Time-varying Operating Conditions Research Articles

Power Allocation Strategy for Urban Rail HESS Based on Deep Reinforcement Learning Sequential Decision Optimization

A hybrid energy storage system (HESS) is adopted to tackle the traction network voltage fluctuation problem caused by high power and large energy demand during the starting and braking of urban rail trains. The system is composed of on-board ultracapacitors and ground lithium batteries, aiming to smooth out the power fluctuation to realize “peak-shaving and valley-filling”. Based on deep reinforcement learning (DRL) online sequence decision, a dynamic power allocation strategy is proposed to improve the energy-saving and voltage stabilization of DC traction networks as well as HESS life protection. Furthermore, to enhance the DRL’s efficiency under time-varying operating conditions, an annealing bias - priority experience replay twin delayed deep deterministic policy gradient algorithm (A-TD3) is proposed to train the replay buffer in DRL. The online learning and optimization strategy is implemented via the mechanism of “trial and error” and “feedback” of the agent. RT-LAB semi-physical real-time simulation systems are adopted to verify the effectiveness of the proposed strategy. Compared with the traditional filtering algorithms and DRL algorithms, the results show that the proposed method converges faster and is more energy saving and stable while effectively protecting the energy storage components.

Cooperative multi-agent deep reinforcement learning based decentralized framework for dynamic renewable hosting capacity assessment in distribution grids

Renewable hosting capacity (RHC) means the total renewable power that can be integrated into the power grid without violation of network constraints. In this paper, a cooperative multi-agent deep reinforcement learning (CMADRL) based decentralized method is proposed to assess the dynamic renewable hosting capacity (RHC) of distribution grids, aiming to duly make decisions for renewable energy interconnection requests and ensure consistent power grids reliability simultaneously. According to the time-varying load-generation operation conditions, the proposed CMADRL method can continuously derive multi-timescale operation strategies for volt-var control devices, e.g., static var compensators (SVCs) and on-load tap changer (OLTC), improving the RHC of distribution grids. With three independent agents (SVC, OLTC and renewable agents), the proposed CMADRL method follows the manner of centralized training and decentralized execution, which guarantees the algorithm convergence under time-varying load-generation uncertainties and meanwhile ensures the feasibility of online applications. The case studies are carried out on a modified IEEE 37-node distribution system to demonstrate the effectiveness of the proposed real-time RHC assessment method. Numerical results verify that the proposed CMADRL method has a better performance than conventional optimization methods on computation efficiency.

A hybrid prognostic method for rotating machinery under time-varying operating conditions by fusing direct and indirect degradation characteristics

Condition monitoring technologies can provide sensor data for predicting the remaining useful life (RUL) of rotating machinery. However, it is often difficult to obtain direct degradation characteristics (DCs), which directly reflect the health of a machine, in real-time. Instead, indirect DCs, which are collected under time-varying operating conditions, are often used. Traditional machine learning and model-based prognostic methods may not be effective when handling such data. This paper presents a hybrid Direct-Indirect Fusion (DIF) method that combines direct and indirect DCs to predict the RUL of rotating machinery under time-varying conditions. The framework can account for time-varying covariates, convert indirect DCs to direct DCs under moderate sample sizes with a loop-generative adversarial network (Loop-GAN), and describe gradual degradation and sudden shocks in direct DCs with Lévy processes. The proposed framework outperforms several benchmarks in predicting the degradation path and RUL of rotating machinery as demonstrated in both simulation examples and an industrial application.

AC Loss Modeling of a Full-Size ITER CS Module

Thorough analyses of electromagnetic transients in superconducting coils for fusion magnets are necessary to compute the losses arising in time-varying operating conditions and correctly design the cryogenic system. The challenge of these analyses, when using numerical methods, is related to the high number of degrees of freedom required to discretize large-scale magnets wound with cable in conduit conductors (CICCs), which include hundreds of composite strands. This study is focused on the central solenoid (CS) of the ITER magnet system, composed of six modules, each wound with several CICC for a total length of 6 km. The numerical code THELMA and an analytical model based on a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">single time constant</i> —previously applied to analyze the tests of a 44-m-long single-layer solenoid, referred to as CS Insert (CSI)—are scaled up to calculate losses during factory acceptance tests (FAT) at the General Atomic (GA) in Poway (US). We demonstrate, by comparison with experimental data, the possibility of predicting the losses in the CS module by projecting the results obtained in the CS Insert subsize test, a single-layer solenoid tested at the National Institute for Quantum Science and Technology (Naka, Japan). The application of a numerical model discretizing the conductor with a few tens of subcables to a full-scale fusion magnet is very demanding. A novel methodology is proposed to abate the computational burden, based on reducing the number of turns involved in the calculation, still retaining a high level of accuracy in the loss assessment.

A physics-informed feature weighting method for bearing fault diagnostics

Intelligent bearing diagnostics has gained popularity over the last few years. However, most of the diagnostic methods are developed under the assumption that training and test data sets are collected under the same working conditions. This assumption is rare in practical scenarios because rotating machinery usually works under wide ranges of rotational speeds and loads. As bearings work under complex and time-varying operating conditions, the test data might come from a data distribution outside the training distribution. Purely data-driven diagnostic models often cannot provide reliable classifications for out-of-distribution test data. To tackle this challenge, this paper proposes a physics-informed feature weighting method for bearing diagnostics. First, a signal processing step is proposed that leverages physical knowledge of bearing faults to extract discriminative features that are robust to bearing speed variation. Then, a novel physics-informed feature weighting layer is developed to assign higher weights for features located closer to bearing fault characteristic frequencies. The feature weighting layer enhances the model’s sensitivity towards the fault-related features among the speed invariant features. Through experiments on three bearing datasets, the effectiveness of the proposed method is validated and shown to have promise for bearing fault diagnostics under different operating conditions. This study also details the deployment of a physics-informed convolutional neural network model on an Industrial Internet of Things (IIoT) device, where edge computing gives users a real-time evaluation of bearing health.

Conditional feature disentanglement learning for anomaly detection in machines operating under time-varying conditions

Anomaly detection(AD) is an important task of machines’ condition monitoring(CM). Data-driven policies can be used in a more intelligent way to achieve anomaly detection and effectively avoid the introduction of expert experience, thus having a broader scope of application. However, Machines like wind turbines often work under time-varying operating conditions(TVOCs), and the performance of traditional data-driven AD methods is significantly degraded because TVOCs can lead to “false alarms” and “missed alarms” in the implementation due to the monitoring data shifting caused by variation of operating conditions(OCs). To address this problem, this paper proposes a novel conditional feature disentanglement learning framework to solve the disturbance in AD on account of entanglement between OCs and health states. The proposed approach performs conditional self-supervised AD by utilizing the variational autoencoder(VAE) and OCs information. Then, a feature disentanglement conditional VAE(FDCVAE) network is developed to realize the disentanglement of OCs and health states. Subsequently, An anomaly indicator(ANI) is constructed by the dimension reduction of the disentangled health state-related feature and combined with the statistics anomaly threshold for AD. Experiments on accelerated fatigue degradation of bearings under TVOCs validate the effectiveness of the proposed method and further demonstrate the superiority of the constructed ANI in eliminating TVOC interference, compared with common fault mechanisms-based and data-driven ANI. The proposed method not only achieves anomaly detection under TVOCs but also provides a new way for representation learning under variable working conditions in machine health management applications in the foreseeable future.

A novel denoising strategy based on sparse modeling for rotating machinery fault detection under time-varying operating conditions

Rotating machinery (RM) such as bearings and gears often operates under time-varying operating conditions (TVOC), which makes the vibration signals non-stationary. In this case, eliminating the effect of non-stationary noise and mining the weak fault information in the vibration signal is the key to implementing weak fault detection of RM. Therefore, a novel denoising strategy based on sparse modeling is proposed in this paper. Firstly, a time series model is utilized to model the non-stationary baseline vibration (BV) generated by healthy RM, and sparse representation theory is introduced to identify the model structure and parameters. Subsequently, a time–frequency response filter is constructed based on the baseline model parameters, which can be utilized to filter out the BV from the raw signal to enhance the fault information. Both simulation and experimental studies verify that the proposed method performs better than several comparison methods in weak fault detection of RM under TVOC.

An Improved Generic Hybrid Prognostic Method for RUL Prediction Based on PF-LSTM Learning

Accurate estimation and prediction of the State-of-Health (SOH) and Remaining Useful Life (RUL) are fundamental to optimal maintenance strategies formulation for Prognostics and Health Management (PHM) of degraded equipment. However, the performance assessment of health state prognostics and RUL prediction is strongly dependent on the errors and uncertainties in physical measurements, and heterogeneous degradation of equipment in time-varying operating conditions. The objective of the paper is to provide a hybrid prognostic framework that integrates a two-phases clustering scheme and a PF-LSTM learning algorithm based on Particle Filter (PF) and Long Short-Term Memory (LSTM) networks for dynamic classification of SOH and long-term RUL prediction in the absence of future observations. The proposed generic hybrid PF-LSTM prognostic approach is demonstrated and compared with other adaptive learning and machine learning methods such as Unscented Particle Filter (UPF) and Radial Basis Function Network (RBFN) on the degradation modeling and RUL prediction for Lithium-ion batteries. The comparison results show that robust prediction performance can be obtained by the hybrid PF-LSTM prognostic approach with accurate characterization of equipment degradation states based on the integrated subtractive-fuzzy clustering analysis. The more accuracy on prognostic estimations in Probability Density Function (PDF) of prior and posterior distributions of battery capacity and RUL that are achieved by particle filtering can gain extensive insights to predictive maintenance action guide.

Mitigating Subsynchronous Oscillation Using Model-Free Adaptive Control of DFIGs

Incorporating subsynchronous damping controllers (SSDCs) into the control loops of doubly-fed induction generators (DFIGs) is one of the most cost-effective methods to suppress subsynchronous oscillation (SSO). However, it is difficult to achieve satisfactory control performance of SSDCs due to the time-varying structure and operating conditions of wind integrated power systems. In this paper, an improved model-free adaptive control (MFAC) method is developed for DFIGs to overcome the limitations of conventional methods for SSO mitigation. First, the structure of the MFAC-based SSDC (M-SSDC) for DFIGs is designed. Then, an improved MFAC predictive control algorithm is proposed to achieve SSO mitigation. Moreover, the stability of the closed-loop system with the proposed M-SSDC is analyzed, which provides some guidelines for determining controller parameters. Additionally, the input-output signal pair of M-SSDC is selected by using geometric measure for the optimum damping performance. Case studies under various operating conditions of the test power system validate the effectiveness of the proposed M-SSDC as well as its superior damping performance over conventional approaches.

Adaptive control of the E. coli-specific growth rate in fed-batch cultivation based on oxygen uptake rate

In this study, an automatic control system is developed for the setpoint control of the cell biomass specific growth rate (SGR) in fed-batch cultivation processes. The feedback signal in the control system is obtained from the oxygen uptake rate (OUR) measurement-based SGR estimator. The OUR online measurements adapt the system controller to time-varying operating conditions. The developed approach of the PI controller adaptation is presented and discussed. The feasibility of the control system for tracking a desired biomass growth time profile is demonstrated with numerical simulations and fed-batch culture E.coli control experiments in a laboratory-scale bioreactor. The procedure was cross-validated with the open-loop digital twin SGR estimator, as well as with the adaptive control of the SGR, by tracking a desired setpoint time profile. The digital twin behavior statistically showed less of a bias when compared to SGR estimator performance. However, the adaptation—when using first principles—was outperformed 30 times by the model predictive controller in a robustness check scenario.

The effect of time-varying operating condition on the crack induced impulses and its application to gearbox tooth crack diagnosis

Gearbox tooth crack diagnosis under Time-Varying Operating Conditions (TVOC) is a challenging issue. TVOC induce both Amplitude Modulation (AM) and Frequency Modulation (FM) effects into gearbox vibration signals, which results in difficulties in tooth crack diagnosis. To overcome this problem, the TVOC-induced AM and FM effects on vibration signals of gearboxes with tooth cracks need to be well studied. Many studies have been reported to investigate the influence of TVOC on gearbox vibration signals using dynamic modelling, some of which only focused on healthy gearboxes while others only considered piecewise constant operating conditions, failing to make comprehensive analyses involving both tooth cracks and TVOC. This paper aims to conduct a comprehensive study on how TVOC affect vibration signals of gearboxes with tooth cracks using dynamic simulation. Firstly, gear tooth mesh stiffness is evaluated considering both tooth crack severity progression and operating condition variations, through which gearbox vibration responses are generated under TVOC. A signal analysis procedure is proposed with its focus placed on the Crack Induced Impulses (CII), which are extracted from gearbox vibration signals using the adaptive harmonic decomposition method or its modified version. Envelope analysis is conducted on the CII to study how the CII are affected by TVOC, and a linear dependence of the AM effect of the CII on TVOC is identified. Based on the identified linear dependence, a novel Condition Indicator (CI), which is sensitive to crack growth while insensitive to operating condition variations, is proposed to track tooth crack severity progression under TVOC. The linear dependence of the AM effect of the CII on TVOC and the effectiveness of the proposed CI for tooth crack diagnosis are demonstrated using both simulated gearbox vibration signals and experimental gearbox vibration datasets.

Research on capacity characteristics and prediction method of electric vehicle lithium-ion batteries under time-varying operating conditions

Electric vehicles (EVs) are being increasingly used today but the safety of their battery system remains an issue. Therefore, a clear understanding of the degradation of EV batteries and the correct prediction of their remaining useful life (RUL) is highly desirable. To solve this problem, firstly, the degradation of EV LIBs at different temperatures and discharge rates are studied experimentally. Then, based on the testing results, the output performance and capacity degradation characteristics of LIBs operating under different conditions are analyze, and a new degradation model is proposed to more accurately predict the capacity degradation of EV LIBs. The research discloses that although the available capacity of LIBs generally decreases with the increase of charge-discharge cycles, at the initial stage of battery use, the available capacity of LIBs are show a transient increase with the increase of charge-discharge cycles. Additionally, the temperature has a significant impact on the available capacity of LIBs, and the alternating changes in ambient temperature can accelerate the degradation of LIBs. Moreover, the innovative degradation model proposed by this paper has higher prediction accuracy than the traditional degradation model, for example, the maximum prediction error given by the proposed model is only 1.17 Ah, which corresponds to a relatively error of 2.34 %; by contrast, the maximum prediction error of the traditional dual exponential model is −5.59 Ah, which corresponds to a relatively error of 11.18 %. These new findings are helpful to the future reliability design of the EVs battery systems.

Attention-based Dual-channel Deep Neural Network For Aero-engine RUL Prediction Under Time-varying Operating Conditions

For complex systems such as aerospace, remaining useful life (RUL) prediction is a general technique that provides information for decision-making in predictive maintenance. In the industrial field, RUL prediction under time-varying operating conditions is a challenging task. In this paper, an attention-based dual-channel deep neural network is proposed to fuse the time-varying operating conditions, with both prediction channels using long short-term memory (LSTM) neural networks. First, the features are extracted by a one-dimensional convolutional neural network (CNN). The operating conditions and sensor data are put into the dual-channel LSTM neural networks separately for prediction. The obtained results are combined with the attention mechanism to assign weights and finally put into the fully connected network for linear mapping to get the final RUL prediction results. This study is based on the N-CMAPSS dataset published by NASA. Compared with traditional methods, this method demonstrates its superiority and effectiveness.

TempGNN: A Temperature-Based Graph Neural Network Model for System-Level Monitoring of Wind Turbines With SCADA Data

Accurate health monitoring and early fault warning are of critical importance to ensure the safe and reliable operation of wind turbines (WTs) and reduce operation and maintenance costs. For this reason, data-driven monitoring approaches have attracted considerable attention and have been widely studied. However, existing methods cannot well consider the interactions of different subsystems and components, thus often leading to missed detections and false alarms. To this end, we proposed a temperature-based graph neural network model named TempGNN to provide system-level monitoring for WTs with the available supervisory control and data acquisition (SCADA) data. TempGNN aims to automatically learn and capture the relationships between sensors from a graph structure data perspective. First, to eliminate the effects of time-varying operation conditions on temperature variables, a decoupled model is designed to obtain operation-independent temperature values. Then, an attention-based adaptive graph structure learning layer is designed to learn the weight matrix, and a spectral–temporal graph network block is introduced to capture spatial and temporal dependencies in graph data. Finally, anomaly detection can be performed through the calculation of health indicators defined with the residuals between the predicted outputs and the actual values. The effectiveness and robustness of the model were verified by two cases on a real SCADA dataset. The results show that our proposed TempGNN model can largely reduce false alarms and provide a reliable monitoring performance.

Health Status Recognition of Rotating Machinery Based on Deep Residual Shrinkage Network Under Time-Varying Conditions

Currently, the research on the health state of rotating machinery under time-varying operating conditions mainly focuses on using a combination of several constant operating conditions or uniformly changing speed and load. This article studied the health status recognition of rotating machinery under nonlinear and continuous changes in speed and load. A health status recognition method of rotating machinery was proposed based on the gram angle field and deep residual contraction network. Considering the influence of working conditions on signal characteristics, the speed, load, and multidimensional time-domain features are fused to form feature vectors. The feature vectors were transformed into images by gram coding. The color contrast relationship mapped from the overall difference distribution of sample feature indexes to the image was not changed while the feature timing was retained, which weakened the influence of working condition information on the sample state, improved the deep residual shrinkage network (DRSN) structure, and introduced the Gaussian error linear unit (GELU) activation function. The experimental verification is completed on the reducer experimental platform and the Xi’an Jiaotong University (XJTU)-Changxing Sumyoung Technology Company Ltd. (XJTU-SY) dataset. The results show that the method can effectively identify the health state of rotating machinery under time-varying working conditions.

Remaining Useful Life Prediction of Ball Screw Under Time-Varying Conditions With Limited Data

Ball screws play a significant role in maintaining transmission accuracy and stability. Accurate prediction of the remaining useful life (RUL) for the ball screw can provide reliable decision-making information for predictive maintenance. However, the RUL prediction of ball screws with the precision indicator under time-varying operating conditions (OCs) is rarely involved. Moreover, obtaining data for the whole life cycle of each OC is impractical due to time and expenditure constraints. To address the aforementioned issues, this article proposes a hybrid approach for RUL prediction under time-varying OCs. First, a time-varying stochastic model considering the precision indicator is established. Then, the data-driven algorithm is proposed to estimate the unknown parameters with limited data. Finally, the effectiveness of the proposed method in prognostics is experimentally validated through the designed platform and compared with the commonly used methods.

AFAFed—Asynchronous Fair Adaptive Federated learning for IoT stream applications

In this paper, we design, analyze the convergence properties, address the implementation aspects, and numerically test the performance of AFAFed. This is a novel Asynchronous Fair Adaptive Federated learning framework for stream-oriented IoT application environments, which are featured by time-varying operating conditions, heterogeneous resource-limited devices (i.e., coworkers), non-i.i.d. local training data and unreliable communication links. The key new of AFAFed is the synergic co-design of: (i) two sets of adaptively tuned tolerance thresholds and fairness coefficients at the coworkers and central server, respectively; and, (ii) a distributed adaptive mechanism, which allows each coworker to adaptively tune own communication rate. The convergence properties of AFAFed under (possibly) non-convex loss functions is guaranteed by a set of new analytical bounds, which formally unveil the impact on the resulting AFAFed convergence rate of a number of Federated Learning (FL) parameters, like, first and second moments of the per-coworker number of consecutive model updates, data skewness, communication packet-loss probability, and maximum/minimum values of the (adaptively tuned) mixing coefficient used for model aggregation. Extensive numerical tests show that AFAFed is capable to improve test accuracy by up to 20% and reduce training time by up to 50%, compared to state-of-the-art FL schemes, even under challenging learning scenarios featured by deep Machine Learning (ML) models, data skewness, coworker heterogeneity and unreliable communication.

Discrete-time contraction-based control of nonlinear systems with parametric uncertainties using neural networks

In response to the continuously changing feedstock supply and market demand for products with different specifications, industrial processes need to be operated at time-varying operating conditions and targets (e.g., setpoints) to improve their economy, in contrast to traditional process operations around predetermined equilibriums. In this paper, a novel control approach is developed for nonlinear chemical processes to achieve time-varying reference tracking, by combining the universal approximation characteristics of neural networks with the rigor of discrete-time contraction analysis and control. This is a robust approach that can deal with bounded parametric uncertainties in the process model, which are commonly encountered in industrial processes. A new synthesis approach is developed that involves training a neural network representation of a contraction metric and differential feedback gain for a full range of parametric uncertainty. This network is then embedded in a contraction-based control structure. A separate neural network is also incorporated into the control-loop to perform online identification of uncertain system model parameters. The resulting control scheme is capable of achieving efficient offset-free tracking of time-varying references under the parametric uncertainty, without the need for controller redesign as the reference changes. Process stability is also ensured during online simultaneous learning and control. Simulation examples are provided to illustrate the approach.

A framework for predicting the remaining useful life of machinery working under time-varying operational conditions

Remaining useful life (RUL) prediction can provide additional capabilities to condition-based maintenance (CBM) and predictive maintenance (PdM) for the reliability and service life of a system. Time-varying operational conditions, such as the altitude, Mach number, and throttle resolver angle of an aero-engine, could result in two main challenges for RUL predictions: varying degradation rates and abrupt jumps in the amplitude of sensor readings. Our study addresses these two challenges in the data pre-processing stage, through operational condition features and the multi-operational condition-based normalization method (MOC-based Normalization). In the framework of our model, first, two density-based clustering algorithms are integrated to be a new classifier for operational conditions clustering and identification in an unsupervised manner. Then, operational condition features consisting of operational condition labels and an operational condition factor are conducted. In the meantime, the proposed MOC-based Normalization recalibrates the upward or downward abrupt jumps of sensor readings at the operational conditions change-points. Sensor data features and operational condition features are combined in the last step of the data pre-processing stage. On this basis, the RUL representation model is trained with the combined features through a gated recurrent unit (GRU)-based network with only two layers in the hidden layer. Experiments on benchmark datasets have been conducted. The results show that the MOC-based Normalization efficiently mitigates the jumps on sensor readings, and the operational condition features improve the prognostic model. Approximately 10% RMSE improvements over the top-three state-of-the-art algorithms are achieved in the RUL prediction under time-varying operational conditions.

Stochastic hybrid system approach to task-orientated remaining useful life prediction under time-varying operating conditions

Mechanical components mainly work under time-varying operating conditions, causing considerable complexity and variability to degradation signals and making it difficult to estimate the remaining useful life (RUL). To handle this problem, this paper proposes a stochastic hybrid system approach to predict the RUL online under time-varying operating conditions by modelling the degradation signals and operating conditions of components. The discrete dynamic nature of operating conditions is described by a continuous-time Markov chain, and their influences on the degradation signals are quantified by degradation rates and signal jumps in a state-space model that is constructed based on nonlinear Wiener process considering the unit-to-unit variability. Considering that a component may have different failure time points under different operating conditions, a novel simulation method based on sequential Monte Carlo is proposed to simulate future operating profiles, predict multiple failure time points and avoid delayed alarms. Hence, the definition of the RUL of the component can be determined through task orientation according to different functional requirements of users, and the accuracy of RUL prediction results can be significantly improved with engineering application value. The effectiveness of the proposed approach is verified via a simulation study and an accelerated degradation test of machine tool spindles.